ShinyButchR

ShinyButchR performs non-negative matrix factorization (NMF) decomposition and signature extraction from genome-scale datasets to identify interpretable features and associate signatures with biological and clinical variables.

Key Features:

• Matrix decomposition: Performs non-negative matrix factorization (NMF) on genome-scale matrices.
• Feature extraction and signature identification: Extracts features and identifies NMF-derived signatures from large genomic datasets.
• Diagnostic plots and visualizations: Generates diagnostic plots to interpret NMF results and assess associations between signatures and biological or clinical variables.
• Integration with ButchR and TensorFlow solvers: Uses the ButchR R package implementing NMF algorithms with TensorFlow-based solvers for optimized computational performance.
• Optimal factorization rank determination: Implements a rational method for selecting the optimal NMF factorization rank.
• Feature selection strategy: Applies a feature selection strategy to prioritize features contributing to identified signatures.

Scientific Applications:

• Signature extraction from genome-scale datasets: Derives interpretable signatures from large-scale genomic data using NMF.
• Association with biological and clinical variables: Associates NMF-derived signatures with biological annotations and clinical variables.
• Gene expression and disease mechanism analysis: Analyzes gene expression patterns to investigate disease mechanisms.
• Identification of candidate therapeutic targets: Highlights signature-associated features that may suggest therapeutic targets.

Methodology:

NMF algorithms implemented in the ButchR package using TensorFlow-based solvers, combined with a rational optimal factorization rank determination method, a feature selection strategy, and diagnostic plot generation for result interpretation.